gru: fix cache coherency issues with instruction retry

Fix two problems related to GRU instruction failures.  Cache coherency is
not maintained for CBEs except when loading or unloading contexts.  When
reading a CBE to extract error information, the CBE must first be flushed
from the cache.

The function that reads kerrnel CBEs was reading the wrong CBE.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This commit is contained in:
Jack Steiner
2009-06-17 16:28:28 -07:00
committed by Linus Torvalds
parent 270952a907
commit 1a2c09e3b4
4 changed files with 14 additions and 4 deletions

View File

@@ -46,6 +46,7 @@
struct gru_blade_state *gru_base[GRU_MAX_BLADES] __read_mostly;
unsigned long gru_start_paddr __read_mostly;
void *gru_start_vaddr __read_mostly;
unsigned long gru_end_paddr __read_mostly;
unsigned int gru_max_gids __read_mostly;
struct gru_stats_s gru_stats;
@@ -376,7 +377,6 @@ static int __init gru_init(void)
{
int ret, irq, chip;
char id[10];
void *gru_start_vaddr;
if (!is_uv_system())
return 0;